The Extreme-scale Scientific Software Stack (E4S) project has released version 22.02 of its collection of software packages for developing, deploying, and running scientific applications on high-performance computing (HPC) platforms.
E4S is supported by the US Department of Energy (DoE) through the Exascale calculation projectand is actually a curated ecosystem of digital libraries, runtimes, and tools intended to facilitate their deployment by the HPC and AI/ML developer communities.
With version 22.02, E4S now includes 100 full-version products, compared to only 24 full-version products and 24 partial-version products for the first version of the collection delivered in October 2018.
The list includes tools like Catalyst, ParaView, OpenMP, Kokkos, and the Flang Fortran compiler, as well as development tools like HPCToolkit, TAU, and PAPI, and math libraries like PETSc and Trilinos.
According to E4S Release Notes [PDF]v22.02 includes a subset of the full Exascale Computing Project (ECP) software technology (ST) product portfolio, but demonstrates a target approach for future delivery of the entire ECP ST software stack.
E4S uses the Spack cross-platform package manager as a meta-build tool to identify and track software dependencies of all packages, and products were apparently targeted for inclusion based on Spack package maturity as well as location in the Scientific software stack, so not all ECP ST packages have been included in this release.
According to E4S project manager Mike Heroux, the project ends up creating hundreds of dependencies so that there is a full software stack available to end users. Heroux is also a principal investigator at Sandia National Laboratories.
“A minimalist way to think of E4S is that it’s a Spack build script, which it really is. Spack is a tool that lets you define how to build your product, and then it lets you ‘identify the other products your product depends on,'” Heroux explained in the E4S Announcement on the EPC website.
“You can think of E4S as the final delivery vehicle for really robust, hardened capabilities that are delivered by ECP in the form of reusable libraries and tools.”
Going forward, E4S will include more and more AI tools and libraries that are specifically needed to apply to scientific problem sets, according to the project team.
The E4S collection is available on GitHub and can be downloaded as a container image in Docker, Singularity, or CharlieCloud format, or via a Spack manifest to install from source.
It is also available as an AWS EC2 image. It supports Arm, x86, and PowerPC CPUs, as well as Nvidia, Intel, and AMD GPUs, so it includes GPU runtimes including CUDA, NVHPC, ROCm, and oneAPI, as well as AI/ML packages like TensorFlow and PyTorch .
Although E4S is open source software released under the MIT license, the packages in the collection each have their own open source license. ®